Analisis Kinerja Proses Pengadaan Dengan Integrasi Process Mining, Bayesian Network Dan Random Forest

Dewi, Ni Putu Cynthia Sasmita (2026) Analisis Kinerja Proses Pengadaan Dengan Integrasi Process Mining, Bayesian Network Dan Random Forest. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6010241018-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (7MB) | Request a copy

Abstract

Pengadaan merupakan proses strategis dalam rantai pasok karena menyerap proporsi biaya yang signifikan dan berpengaruh terhadap efisiensi operasional perusahaan. Salah satu proses penting dalam pengadaan adalah Purchase-to-Pay (P2P), yang dalam praktiknya memiliki berbagai variasi alur proses, rework,serta bottleneck. Penelitian ini mengembangkan kerangka predictive monitoring berbasis event log dengan mengintegrasikan Process Mining, Bayesian Network, dan Random Forest. Data yang digunakan merupakan event log proses P2P yang terdiri atas 16.190 cases, 103.799 events, dan 23 aktivitas. Process mining digunakan untuk menganalisis alur proses aktual, bottleneck, rework, serta throughput time. Bayesian Network digunakan untuk membangun probabilitas terkait aktivitas yang akan terjadi berdasarkan aktivitas yang sudah terjadi, sedangkan Random Forest digunakan untuk memprediksi durasi antaraktivitas. Hasil penelitian menunjukkan bottleneck utama terjadi pada transisi Record Invoice Receipt menuju Clear Invoice sebesar 44,82%. Rework tertinggi terdapat pada Change Quantity sebesar 18,14% dan Change Price sebesar 14,16%. Bayesian Network yang dibangun terdiri atas 32 node dan 65 arc, dengan rata-rata log-likelihood per node sebesar -0,08360 pada 5-fold cross-validation. Random Forest menghasilkan R² data uji sebesar 0,7274, MAE sebesar 4,59 hari, dan RMSE sebesar 7,37 hari. Integrasi Bayesian Network dan Random Forest kemudian diimplementasikan pada case yang belum mencapai tahap akhir yang kemudian menghasilkan rata-rata probabilitas Top-1 path sebesar 0,8994 dan rata-rata prediksi sisa durasi sebesar 37,85 hari. Sebanyak 99 case atau 19,80% berada pada kelompok durasi tinggi. Hasil ini menunjukkan bahwa integrasi Process Mining, Bayesian Network dan Random Forest dapat mendukung monitoring incomplete case melalui informasi aktivitas apa yang akan terjadi selanjutnya serta prediksi sisa durasi hingga mencapai tahap akhir.
========================================================================================================================================
Procurement is a strategic process in the supply chain because it accounts for a significant proportion of costs and affects the operational efficiency of a company. A key component of procurement is the Purchase-to-Pay (P2P) process, which can involve various flow variations, rework, and bottlenecks in practice. This research introduces a predictive monitoring framework based on event logs by combining Process Mining, Bayesian Network, and Random Forest. The study utilizes a P2P event log comprising 16,190 cases, 103,799 events, and 23 activities. Process mining is employed to examine the actual process flow, identify bottlenecks, rework, and measure throughput time. The Bayesian Network is utilized to estimate the probabilities of future activities based on past occurrences, while Random Forest predicts the duration between activities. Findings reveal that the primary bottleneck is in the transition from Record Invoice Receipt to Clear Invoice, representing 44.82%. The most significant rework occurs in Change Quantity at 18.14% and Change Price at 14.16%. The Bayesian Network constructed includes 32 nodes and 65 arcs, with an average log-likelihood per node of -0.08360, validated through 5-fold cross-validation. Random Forest achieves a test R² of 0.7274, MAE of 4.59 days, and RMSE of 7.37 days. The integration of Bayesian Network and Random Forest is applied to cases not yet completed, yielding an average Top-1 path probability of 0.8994 and an average predicted remaining duration of 37.85 days. In total, 99 cases, or 19.80%, fall into the high-duration category. These findings suggest that combining Process Mining, Bayesian Network, and Random Forest can aid in monitoring incomplete cases by indicating the next likely activity and estimating the remaining time until completion.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Purchase-to-Pay (P2P), Process Mining, Random Forest, Bayesian Network, Predictive Monitoring, Purchase-to-Pay (P2P), Process Mining, Random Forest, Bayesian Network, Predictive Monitoring
Subjects:	H Social Sciences > HD Industries. Land use. Labor > HD39.5 Industrial procurement. Q Science > QA Mathematics > QA279.5 Bayesian statistical decision theory. T Technology > TJ Mechanical engineering and machinery > TJ217.6 Predictive Control
Divisions:	Faculty of Industrial Technology and Systems Engineering (INDSYS) > Industrial Engineering > 26101-(S2) Master Thesis
Depositing User:	Ni Putu Cynthia Sasmita Dewi
Date Deposited:	16 Jul 2026 03:38
Last Modified:	16 Jul 2026 03:38
URI:	http://repository.its.ac.id/id/eprint/135135

Actions (login required)

View Item